Healthcare Unbound_Todd Gottula: this mp3 audio file was automatically transcribed by Sonix with the best speech-to-text algorithms. This transcript may contain errors.
Intro/Outro:
Welcome to Healthcare Unbound, a podcast powered by Clarify Health, where healthcare's changemakers discuss ways to advance care outcomes, cost, and affordability.
Saul Marquez:
Hello everyone, and welcome back to the Healthcare Unbound podcast, brought to you by Clarify Health. I'm super excited to be hosting someone that you've heard of on the show before. His name is Todd Gottula. He's the co-founder and president of Clarify Health Solutions, based in San Francisco. And as the co-founder and president, he believes that technology and clinical expertise can improve the lives of patients and caregivers. They're doing this in a big way, leveraging data, as well as the Atlas platform that maps over 300 million patient journeys to deliver 18 billion-plus AI-predicted precision insights. They've been recognized by KLAS research, the last one in 2024, as in the best-in-KLAS awards for software and services. And they're really just receiving all sorts of recognition. It's a lot of wows here. And Todd, I'm excited to welcome you back to the podcast. Thanks for joining me.
Todd Gottula:
Thank you so much for having me. I'm looking forward to the conversation.
Saul Marquez:
Likewise, and for everybody that hasn't had a chance to listen to our last conversation, we'll reference it in the show notes. Todd dug deep into AI, and there's been some real advances even since our last talk in artificial intelligence and especially in healthcare since we last spoke. So, Todd, why don't we just start there as a good continuation of where we left off?
Todd Gottula:
Yeah, it's been an exciting, I think it's probably been almost 10, 11 months since we last spoke, and I'd say that there's been two pillars of real advancement. One is the underlying capabilities that companies like Clarify can take advantage of and how those enabling capabilities have advanced so significantly, which I'll speak to in a moment, and then the second is really the crystallization of what the applications of these technologies can be to our space, and I want to talk about that first. And really, I'd split that further into two categories. One is what I like to call conversational user experience. So this is affording everyone in the healthcare ecosystem the opportunity to interrogate and interact with information that has historically been Tabular, Excel, Tableau, grid-like user experience, right? We're all familiar with this for the last 2 or 3 decades of how you interact with information. Now, with a conversational user experience, you can talk to the data and you can start to interrogate it with familiar words and phrases, just making that information much more accessible and available. And then the second is what I like to call the needle-in-a-haystack problem. Although we talk a lot about generative AI being about large language models and thinking of it as really being text-based, what also is happening there is just by the way that these technologies work, you're able to index and create large amounts of information in incredibly new and novel ways that affords a retrieval of anomalies in an efficient way that's never been seen before. So broad strokes, you've got technology advancements, you've got our application of those. Under our application, you've got this conversational user experience, which is really evolving to be something that's quite powerful. And then you've got this incredible needle-in-a-haystack capability that is allowing us to more rapidly find what matters in these underlying data sets. You put that all together, and we are now at the place where it's not just a frontier or an evolution, but we're at the place where we're starting to realize real impact, and we're seeing direct applicability of these sort of emerging and sometimes, Hey, how is this ever going to apply to healthcare type of technology? And ten months later, we're like, it's here, we're implementing it, we're using it every day, we're making it available to our customers, and it's just an incredibly exciting time.
Saul Marquez:
Yeah, I couldn't agree with you more, Todd. I was just chatting with some colleagues around gen AI and its impact on search. Like, who would have thought search would be impacted by it?
Todd Gottula:
Yeah, it's very similar to what I was speaking around this needle-in-a-haystack problem. Because when you're doing the search, what are you trying to find? You're trying to find the thing that matters most to me that's out there in an ecosystem of information or in a lake of data, and being able to both find that information but also understand the context around your question, such that what is brought back to you is actually what you want, as opposed to a list that might be stacked ranked based on who's paying for you to see that more than somebody else. That's really what's so exciting. And you're right, that parallel between traditional Google or Bing searching, now think about that in healthcare, where we've got this large amount of historical encounter information. Like a bunch of stuff has happened, we've organized it in a meaningful way. So what do I do with that? How do I make sense of it? What am I supposed to do next? Those types of higher-order questions are now answerable with these technologies put on top of the large data sets like the ones that Clarify have created.
Saul Marquez:
It's so powerful. And I think that's the, during our last conversation, the one thing that you said that really stuck with me was ChatGPT, the fact that there's commercial availability of these technologies has really made it applicable to everyone. So why don't we, why don't we dig into what natural language interface is, and specifically how we could be thinking about things that are commercially available to healthcare and how can we enhance patient care with that?
Todd Gottula:
So one of the most enabling and impressive technology evolutions that's really happened over the course of the last 6 to 12 months is the standardization of retrieval augmented generation, RAG, approaches because one of the historical concerns about these large language models was, I have to contribute a lot of healthcare data that either is governed by HIPAA or whether, maybe it's de-identified, but we still wouldn't be comfortable co-mingling it with the large training data sets that have been aggregated from publicly available documents. And so there's been these concerns like, how would we possibly be able to enable a generative AI model on top of healthcare data? Now there's technologies that enable that in a much more robust and meaningful way, where you can have a private data asset that is augmenting one of the more commercially available large language models to be able to get the best of both worlds. And really, it's been light speed with the maturity of evolution, or the pace at which those technologies have become mature, and the standardization of the packages so that we can start docking those things in without having to spend my engineers' time, which is not their background and expertise doing that type of coding. We can be doing integration development, so that's one aspect. And then the second is just, I guess arms race is, I don't know, it's the competition that has evolved, and everyone's in, right? You've got obviously OpenAI, you've got what is going on with Amazon and with what Facebook is doing, and just across the marketplace of with Anthropic and Claude. Like, everyone is spending billions of dollars developing out these technologies and then making it available for a fraction of what it costs for them to develop it for organizations like us, and...
Saul Marquez:
So, is it data? Are they doing this for data or just a market grab?
Todd Gottula:
I, it's a great question. And I'm obviously not an expert sitting in their boardrooms, but I think there is a hypothesis that much like Google's second and third mover place to dominance in traditional web search. Although I'm sure Microsoft would not like me saying that. I think there is a hypothesis that there will be a market-dominant provider of these capabilities, and that will be a significant market share, and therefore, there'll be opportunity to take price and be ubiquitous across the landscape, and so therefore it's worth the investment. Yeah, it's a land grab. And we all stand to benefit, right? I have, I read something recently that the entire Saturn moon landing program, less was spent in today's dollars on that than what one of these organizations is, are spending right now on compute power, actual electricity, and engineering to be training these types of base models that we get to take advantage of.
Saul Marquez:
Yeah. It's huge, it's huge. It's a new world for sure, and it's a new world for healthcare as well with what you brought up around being able to merge these data sets without worrying about putting the private data that you have on the outside. What are the key benefits of being able to interact with healthcare data systems as if conversing with a knowledgeable advisor?
Todd Gottula:
Yeah. So, think about this problem that I'm sure everyone has been presented with, and those of us in the healthcare industry get presented with more than most, which is I need to go to a doctor. Can you tell me the best doctor to go see? So that question is actually an incredibly complex question to answer because it's inclusive of: what care do you need? What are your existing conditions? What providers of service are available in your marketplace? Who covers or who carries your insurance? How much is that going to cost you with your insurance or without, and who's available? And can that individual actually see you? So, that problem of provider availability and patient-provider matching has been one that the healthcare industry has struggled with or has been trying to solve since inception, really. And there's lots of different ways that people have tried to come at this by defining networks and narrow networks and saying, Hey, I'm going to select some docs, and any one of those docs is going to be right for you, but honestly, that's not always the case. So what we're now able to do is put that conversational user experience on top of all of the incredible provider performance data that we have at Clarify, which isn't just this physician is great at doing total joint replacements. It's more granular than that. It's this physician is great at total joint replacements for these types of individuals, and this physician serves Cigna commercial patients with low copays and deductibles and typically is able to see that patient within 30 days. So that data lake that we've created, now we put a conversational UI on top of it. And now we can put in the hands of a primary care physician, a care coordinator, or potentially in the future, maybe even the general public, the ability to come in and say, like my mom asked me two weeks ago that she needed to go see somebody for chronic heartburn in her area, so she needed to see a GI specialist in southern Oregon that takes Medicare, and her doctor's referral was four hours away. It turns out I was able to find someone who's a really high-performing person, particularly specializes in Medicare-aged women who suffer from this, which is actually a quite typical condition, and she was able to see that physician in like 18 days, and she's got great care.
Saul Marquez:
That's awesome. It makes a difference, and this is a great example of what can be available today. And let's unpack that a little bit more, Todd. So with AI-driven physician selection, like how does this process work and what specific criteria are considered when recommending physicians? I know you've dug into some of it, but like let's get more granular here.
Todd Gottula:
Yeah. What, maybe be helpful for me to walk through what we've done here at Clarify to lay the baseline, because that creates the, let's call it the private data asset that then is used in concert with the base models that are available from OpenAI and Anthropic, which are the two partners that we use. So we've taken a large amount of de-identified, HIPAA-compliant historical claims data. We've then converted that into billions of patient journeys. So think of a patient journey as an individual went to a hospital, and we're looking at all of the care that they received, all of their outcomes, the quality associated with that, and the cost. And all of that is tagged back to or linked to a provider, a physician, or, and/or the side of service, the hospital where the surgery or the event occurred. So, we have this deep data lake where all this information exists. Now, because we have that, we can then start to identify positive and negative variation, strength and weakness of care patterns, what providers or service, what facilities or doctors tend to have better outcomes for what types of patients. So that's what we have. And then we use a RAG implementation to basically say, Look, let's take a base model. Let's break that down into a series of, our data assets into a series of agents that understand proximity to the patient, what plans that physician actually covers, what type of services they offer, and then the quality of their services. And then we run an orchestration to be able to say, Okay, Todd is asking for a physician for a patient that matches this characteristic for this type of service in this geography. And then we go through and we have an orchestration that basically will surface back the most likely physicians that would be appropriate, along with their contact information and availability for the question being asked. And again, I know I went through a sort of technical breakdown of that, but.
Saul Marquez:
No, that's great.
Todd Gottula:
That process is something that we couldn't do 12 months ago, right? The technology wasn't there. The, we hadn't organized our data or realized how we had to organize our data. And then the base models hadn't matured to be able to understand healthcare context. That's another really important part, is that there's been a lot of investment such that the OpenAI and Anthropic, specifically base models, can understand and provide to us the ability to do recall from embeddings on medical terminology so that somebody actually comes in with a more technical definition of heartburn or total joint replacement, and they actually use the healthcare technical terms for those that we're able to understand that.
Saul Marquez:
Man, that's fascinating. And two things that come top of mind, Todd, is cost. Is that a factor that could be like, you're talking value-based care: Is cost something that could be factored into this search?
Todd Gottula:
Absolutely. So, we'd like to use the word affordability as opposed to cost because affordability is the important measure. Because what you're doing there is you're assessing the true relative expense to both the, whomever is paying for the services themselves, whether it's an insurance carrier or an employer and the employee, and both of those things matter, or the patient. I've said that in an employed context, and so we're able to have that information available in our data lake. And yet you can ask questions of, Look, I want an adequate provider who is the lowest cost to the patient. And those are criteria that our conversational user experience can now factor in and surface a different set of physicians if that's what you're looking for, as opposed to cost don't matter; I just want the best. Or I want the person who can see me tomorrow because I am in such need of services, I don't really care that much as to what my out-of-pocket spend is going to be for that care. And so again, this kind of points to the complexity, to the answer of the question of tell me which doctor I should go see and why these technologies are so enabling and empowering because now we can actually take all that into context and give individuals the answer they're seeking because they can just speak it, they can just type it, as opposed to having to sift and sort through a whole bunch of Excel sheets and tabular data.
Saul Marquez:
That's awesome. I love that breakdown. It makes it so easy to understand, Todd. Thank you for that. So it's crazy, right? Because we were talking 11 months ago and so much has changed. I think you and I need to be meeting here at least every six months.
Todd Gottula:
I love that. I love talking to you.
Saul Marquez:
Man, because this is too much time.
Todd Gottula:
I'm sorry. Can I give you one more example ...?
Saul Marquez:
Please, please, please. I love doing this, yeah. I'm sure our listeners are too, yeah.
Todd Gottula:
The price transparency data, right? That was the congressional mandate that both health systems and health plans had to publish the negotiated rates that they have for service costs. We pull all that data in and in aggregate across roughly 100 health plans, and, I think, about 3000 hospitals; it's six petabytes of data. So yes, this data is technically transparent, but another really exciting use of these technologies that we've been discussing here is to be able to find the needle-in-a-haystack about where are there rate outliers, where someone is not being paid fairly, either being overpaid or underpaid for a service in a market which ultimately impacts all of us, right? The whole idea behind the Price Transparency Act was to level the playing field to illuminate the market is paying for the same service, and then organizations like us overlaying quality to be able to say, Yeah, this rate actually is justified being hired because it's higher quality outcomes. Generative AI, on top of the rate data, has proven to be an incredible unlock because now, again, going back to conversational user experience on top of a large data lake, being able to find the needle in a haystack, to say this is the rate that you should go. Either as a health plan, you should go to this provider and say, Hey, look, we need to talk about this. Or as a provider going, Hey, I'm delivering high-quality services, and I'm getting underpaid; you should pay me fairly. It's an incredible unlock for the industry at large, ultimately meaning that consumers like us who pay for healthcare are going to benefit.
Saul Marquez:
Totally. No, I love that. Thanks for that example. And even we've talked a lot on our podcast about value-based care. I'm sure there's a lot of unlocks there and maybe we could make that a follow-up.
Todd Gottula:
It could be. Look, I'll just, I'll plant a seed for you, which is I think that the words value-based care have unfortunately become so tainted in this industry. I like to say now that we need to create the next generation of fee-for-service that actually delivers better patient outcomes, and I believe that there is a way to ride the fee-for-service chassis that this entire industry is built on and will not change to actually create variable rate structures that reward those that should be rewarded, which are the providers who are doing the actual healthcare delivery in this country will higher quality work that they're performing. That is the true way for us to deliver better outcomes, which is where all of us want out of this system, but to do it in a way that finally breaks down this paradigm or the resistance to value-based care or at-risk models. So you're absolutely right. These technologies are a key part to us creating this next generation of fee-for-service, which I think is going to be the way that we finally break through and revolutionize the way in which care is delivered in this country.
Saul Marquez:
I love how you phrased it: the next generation of fee-for-service. I think that strikes a better chord. I appreciate you sharing that. All right, Todd, it's time for you to bring out the crystal ball, okay? And what are your predictions for the future of AI in healthcare, particularly in the context of physician selection and patient care?
Todd Gottula:
So organizations like Clarify, and I would like to say, led by Clarify, will start releasing more AI-enabled agents into the marketplace for broad use by health plans, health systems, and digital health companies, thereby unlocking another stairstep advancement in how services, capabilities, and offerings can be developed and deployed. What I mean by that is, if you think historically that, we talk about this API, a programmatic interface to a data set where someone can dock up a program, ask a question, get a response. Those are no longer going to be restricted to coders. Those are going to be now enabled for much higher order type question response, and those things can be linked into orchestrations that are going to solve very complex problems. There's a classic demo that OpenAI has of an individual saying, I want to go to Iceland, this is how much money I want to spend, build me a trip. And multiple agents are used around airfare, hotel, understanding what availability and events are, and they build an itinerary and they can go so far as to build and book the flights, book the itinerary, schedule the hotel, etc.. Imagine a world where that's able to be orchestrated on a set of underlying agents in the healthcare context. Evaluate the physician, schedule the appointment, book Uber Health to get transportation for the patient to get there, and then make sure the follow-up occurs and maybe it's a digitally enabled follow-up, all of those things orchestrated by one party, but you leveraging the services of multiple underlying healthcare individuals. So it's a long way of saying, I think this is going to cause a real dramatic unification of the underlying capabilities that sit in healthcare latent today, ultimately to a much, much better patient experience, and ideally, again, better patient outcomes, which is what we're all here for.
Saul Marquez:
Totally agree. By the way, I had no idea that they could actually book your flights for you. Can it actually get that done?
Todd Gottula:
It can. Although I think the technology is still evolving because there's a great demo, but then when people have tried to do it, it's a little.
Saul Marquez:
Hey, but listen, 11 months ago, this stuff wasn't available.
Todd Gottula:
Exactly, exactly. That's the thing, that light-speed pace at which These technologies are evolving and then getting implemented. We're not even talking about on the video and audio side. There are so many things that are happening so quickly, it's really hard to say you'd have a crystal ball and be anywhere near, but how are you also going to be accurate with the pace of evolution right now?
Saul Marquez:
Oh man, for sure. And I'll tell you, Todd, I was, and everybody listening, I was, I changed it up. I usually go to healthcare conferences. I went to a customer experience call center conference. And I saw what they're using gen AI for. They're literally able to monitor 100% of call center conversations and improve experience. It's just incredible.
Todd Gottula:
It's absolutely incredible. Like the Salesforce Einstein enabled, I guess it's not called Einstein anymore, but the ability to, while you're having a conversation in line with an individual, have additional information brought in about who you're talking to, not just about what products they use, but who they are as people and what were their most recent posts on various social media platforms, it's, it just provides this encompassing experience. And again, as we've been talking about, I think we're just at the tip of the iceberg.
Saul Marquez:
Totally agree. Todd, I got one last question for you. How can healthcare providers, and patients, and payers stay updated on the latest advancements in AI-driven healthcare technologies?
Todd Gottula:
Look, it's a great question. I spend a lot of time across all the various information sources. Believe it or not, Twitter's got a bunch of different forums and interesting individuals to follow. Just watching what's happening with the AI and Anthropic and reading their blogs about just the development of these technologies, and then staying plugged into individuals like yourself who are reporting on this industry. It's a lot of information. There's a massive firehose of knowledge that's out there. But I'd say the other way is that, come to organizations like us, but really ask your existing service providers where they are, what's next, what's coming in their roadmaps, and really stay plugged in into this industry at large.
Saul Marquez:
Love it. Some great advice there, Todd. Thank you so much always for bringing your A game and your expertise to the podcast. Folks, for all of the show notes on today's episode with Todd Gottula, check out the show notes and make sure you follow us on social and comment in the LinkedIn posts that we will be posting about this episode. We want to hear from you. We want you to be part of the conversation. Todd, thanks so much for being with us today.
Todd Gottula:
Always a pleasure. Thank you very much, Saul.
Intro/Outro:
Thank you for listening to Healthcare Unbound. We hope today's episode was insightful. If you want more information on how Clarify Health can help you, please visit ClarifyHealth.com.
Sonix has many features that you'd love including enterprise-grade admin tools, automatic transcription software, advanced search, transcribe multiple languages, and easily transcribe your Zoom meetings. Try Sonix for free today.
